DDIB: Dual Diffusion Implicit Bridges for Image to Image Translation

**Is your feature request related to a problem? Please describe.**
Common image translation methods typically depend on joint training using data from both the source and target domains. However, [DDIB](https://suxuann.github.io/ddib/) addresses this challenge by independently training diffusion models on each domain, while maintaining a shared latent space across all domains.
The original DDIB code lacks optimization, leading to lengthy processing times. Additionally, unlike diffusers, it presents challenges when attempting to integrate with other pre-trained models or schedulers.

**Describe the solution you'd like.**
An implementation of DDIB with diffusers that can be run with HF accelerate could speed up the operations and will allow to work more easily with other implementations of models and schedulers on diffusers.

**Describe alternatives you've considered.**
Tried to execute the original code but is too tedious.

**Additional context.**

Paper: https://openreview.net/forum?id=5HLoTvVGDe
Github: https://github.com/suxuann/ddib
Project Page: https://suxuann.github.io/ddib/